KPMG's Q2 2026 Global AI Pulse surveyed 2,145 senior leaders across 20 countries, and the headline stat is already everywhere: only 7% of organizations have established real ROI on their AI investment.
Dig one layer in, and the visibility numbers explain why. Only 35% of leaders have full visibility into what AI actually costs to run. 23% say they can't get a handle on usage-based costs. 42% call their visibility partial at best. Leaders with strong cost visibility are five times more likely to report established ROI (15% versus 3%).
The Cost Side Has a Mirror Image
Every statistic in the KPMG survey has a revenue-side counterpart that nobody is publicly measuring. If 65% of organizations cannot fully see what AI costs them internally, how many can say with confidence that every unit of AI-driven usage a customer consumes is priced, billed, recognized, and reported correctly on the way out the door?
Internally, a cost visibility gap means finance can't trace a number back to its source. Externally, a revenue visibility gap means a customer's invoice might not match what they used, recognized revenue might not match when value was delivered, and a renewal conversation might be built on usage data nobody double-checked.
Traditional software gave finance a fairly stable floor to stand on. Licenses got negotiated, contracts got signed, and the number on the invoice didn't move much once the deal closed.
AI pricing breaks that stable floor. Costs move with usage. Demand shifts week to week. Pricing varies by model and by customer. Value gets delivered continuously instead of on a renewal date.
Finance teams are being asked to run commercial models their systems were never built to support, and that's true whether you're looking at the cost ledger or the revenue one.
Why the Revenue Side Is Harder to Ignore
A cost visibility problem can sit quietly inside a company for a while. Nobody outside finance necessarily notices if the internal AI spend dashboard is a rough estimate instead of a precise number.
A revenue visibility problem doesn't get that grace period. It shows up the first time a customer disputes a bill, the first time a renewal negotiation stalls because nobody can explain how a usage-based price was calculated, or the first time an auditor asks how recognized revenue ties back to actual AI-driven usage instead of a flat subscription fee.
This is the same test SaaS billing failed early on, just compressed into a shorter timeline. Consumption is arriving faster than the systems built to meter, rate, and report on it. The companies that are struggling are the ones whose commercial plumbing was built for a business model that has since moved on without it.
What Revenue-Side Readiness Actually Looks Like
Can you trace a dollar of AI-driven usage from contract to billing to recognized revenue without someone stitching systems together by hand? When usage swings hard month to month, does the forecast move with it, or does finance find out three weeks later?
Answering yes to both takes the same discipline any usage-based model needs: usage captured at the source, rated against a defined pricing model, billed accurately at any volume, recognized when value is actually delivered, and reported through a system built to hold up in a board meeting; not assembled the night before one.
The Question Worth Asking Alongside KPMG's
KPMG's data pushes finance leaders to ask whether they can see what AI costs them. The harder, less-asked version of that question is whether they'd trust their own numbers on the revenue side if a customer, an auditor, or a board member asked the same thing tomorrow.
If the honest answer is "not yet," you have a revenue architecture problem, and unlike the cost side, it's usually the customer who finds out first if the answer stays unresolved.
At Synthesis Systems, we help revenue and finance leaders build the metering, billing, and reporting architecture that usage-based models, AI included, need on both the cost side and the revenue side. That work typically starts with Synthesize, our structured assessment of the systems and data connecting usage, cost, and revenue. If that sounds familiar, contact us below to talk about a Synthesize assessment, or email Sales@synthesis-systems.com.